Primer’s Sean Gourley: “AI is the biggest change to war since the internal combustion engine”
发布时间 2023-03-17 04:29:41 来源
摘要
The Sunday Times’ tech correspondent Danny Fortson things on Sean Gourley, founder of Primer AI, to talk about how he got into the industry (3:45), and then getting into counterintelligence (6:40), the image recognition revolution (9:00), raising money from the CIA (11:00), AI in war (16:30), how machines beat human Top Gun pilots (20:00), AI as the “third offset” (22:00), how the US is still living with a Cold War mentality (27:30), the AI arms race with China (30:00), the long ties between the Pentagon and Silicon Valley (34:15), how it has become easier to recruit (37:30), the Chat GPT effect (40:40), the next weapon of mass destruction (45:00), and Tiktok and the information war (48:00). Hosted on Acast. See acast.com/privacy for more information.
GPT-4正在为你翻译摘要中......
中英文字稿 
I'm Alex Dibble from The Times. There's so much happening in the world right now. How do you stay on top of it? On the World in 10 podcast, near my team, sit down and go through it all in just 10 minutes. The Times correspondents join us with eyewitness accounts and interviews. I was actually on the orthopedic ward, almost all injuries are shelf-fired. A woman who was inside the police van. They also give us their unique analysis of world events. Give the world in 10-a-listen, it is 10 minutes to stay on top of the world.
我是时报的Alex Dibble。现在，世界上发生了很多大事。您是如何紧跟最新消息的呢？在“世界十分钟”播客中，我和我的团队，仅用十分钟的时间让您了解所有最新动态。《泰晤士报》的记者与我们一起分享他们的亲身经历和采访。我曾在骨科病房，几乎所有的受伤都是自己造成的。其中一个女人曾经被关在警车里。他们还将提供他们独特的世界事件分析。请花10分钟时间聆听“世界十分钟”，以保持对全球的关注。
Yo! Hickmology. What is it all about? One of the latest ones was, can a machine pilot, and I think there were 16s, but I don't know, maybe there were 20 toes. Pilot a plane better than a human top gun. And they ran this. There were five, I think five or six teams that entered. The winning team that went against the human, so it was a human top gun and a simulator against a machine. In a dog fight. In a dog fight, one on one. Right, one on one. The machine beat the human five zero, not even close. One might think about it as autopilot steroids. And it's got so good now that it's better than the best human. And the next step on that. So Maverick and Goose are screwed. Yeah, the Goose is cooked, right?
嘿！“Hickmology"是什么意思？其中一个最新的问题是：机器能比人类的“顶尖飞行员”更好地驾驶飞机吗？我记得有16个团队参赛，但我不确定，也可能有20个脚趾。他们进行了比赛。有五个团队参加了比赛，对抗人类“顶尖飞行员”和模拟器，与一台机器进行了一对一狗斗。机器以5比0的成绩击败了人类，结果非常明显。有人可能认为这就是自动驾驶的增强版。现在它已经非常好了，比最优秀的人类驾驶员更好。下一步应该是什么呢？所以，雄鹰和欧鸟要完了。是啊，欧鸟煮熟了，对吧？
Hello and welcome to Danny in the Valley or weekly dispatch from behind the scenes and inside the minds of the top people in tech this week. We're back with more AI stuff after a little detour to human composting last week. Hope you guys enjoyed that one. But anyhow, really, is there anything else going on in tech aside from AI? You know, what is crazy actually between last week's recording and this week's episode we had a bank run a bailout. We've had the White House reportedly telling bite dance, the owner of TikTok that they have to sell the app or get banned in America. And then of course, there's also the release of GPT 4, which is a dramatically more powerful version of chat GPT, which all works out great because this week's guest is on to talk all things AI and specifically what it means for defense and how the technology could really change war and conflict forever and really in unimaginable ways.
大家好，欢迎收听 Danny in the Valley，这是我们每周的专题节目，从科技行业内部和顶尖人物的角度，深度探讨各种话题。这周我们又回归到 AI 的话题，上周我们稍微有点偏离主题，聊了一下人体堆肥，希望大家喜欢。不过其实除了 AI，科技行业还有其他的动态吗？很疯狂的是，在上次录制和这一次播出之间，我们经历了一家银行的紧急救市，白宫据称要求 TikTok 的所有者 Byte Dance 出售该应用程序，否则将被禁在美国。当然，此外还有 GPT-4 的发布，它是一个更强大的 GPT 版本，这一切都非常棒，因为我们本周的嘉宾将谈论 AI 的所有事情，特别是它对国防的影响，以及这种技术将如何彻底改变战争和冲突，以及未来可能发生的难以想象的变化。
So this week's guest is Sean Gourley. He is the chief executive and founder of primer AI, a security company that develops AI tools for intelligence and the military establishment. Gourley is a serial entrepreneur. He's been at the forefront of big data and AI for years. He has a really unique vantage point, especially as these tools like GPT 4, which can pass virtually every standardized test, write complex software code, turn some scribbled handwritten instructions into a website. These tools are really setting the world light, but for defense, what they show is just this rapid development of AI, means something very different, especially vis-a-vis the West competition for supremacy with.
这周的嘉宾是肖恩·高利。他是Primer AI的首席执行官和创始人，这是一家为情报和军事机构开发AI工具的安全公司。高利是一位连续创业者。多年来，他一直处于大数据和人工智能的前沿。他有非常独特的视角，特别是像GPT 4这样的工具，它可以通过几乎所有标准化测试，编写复杂的软件代码，将一些潦草的手写指令转化为网站。这些工具正在改变世界，但对于防御来说，它们展示的人工智能的快速发展意味着非常不同的事情，特别是在西方争夺霸权的竞争中。
So I admit all the hullabaloo Gourley can just give a very different perspective on how we should be thinking about these momentous times in which we live these incredibly powerful technologies which feel like they're just kind of coming out of nowhere. And just before we get started, very important programming note, we had this conversation about 10 days ago. So before the news emerged that the White House hasn't reportedly ordered the sale or ban of TikTok in America. So just when we talk about that, please bear that in mind. It doesn't make the conversation any less pertinent, but just for time being perspective, just keep that in mind.
我承认，高利制造的喧嚣可能会给我们带来完全不同的视角，帮助我们更好地思考我们生活中这些重要时刻和那些看似突然出现的、异常强大的技术。但在我们开始之前，我想特别提醒一点：我们是在10天前进行了这次对话，因此在我们谈论部分内容时请注意，最近白宫曾被报道要求出售或禁止抖音在美国的情况还未出现。这并不会使我们的讨论变得不重要，但是请在此期间持有正确的视角。
So anyhow, this is a good one. So I'm going to stop talking and hand over to my conversation with Sean Gourley of primer AI. Enjoy.
总之，这是个好节目。我现在要停止讲话并把话题转交给primer AI的Sean Gourley。请享受本次对话。
Chad G.P.T. kind of comes out, what, November? Yeah, I think it was started in November, right? Yeah. And that has kind of set the world light in a lot of different ways. And in this past month, I've been kind of going around the house, talking to various different people in some aspect of AI. I was just had a really interesting conversation with Stuart Russell at UC Berkeley who's kind of written a textbook. The most pirated textbook in artificial intelligence, he will tell you. Exactly. He's, as I'm sure you know, really worried about autonomous weapons. And also autonomous weapons and more broadly, this idea of how do we build AI that we can still control? In other words, that doesn't kind of supersede us and figure out that it wants to do things that are not beneficial to humans. So I was really interested in conversation, but along my travels that came across you guys, and I'd love to just get a sense.
查德G.P.T.是在11月份发布的，对吧？它在很多方面都引领了这个世界。最近一个月，我一直在家里走来走去，跟各种涉及AI方面的人交谈。我刚和加州大学伯克利分校的Stuart Russell进行了一次非常有意思的对话。他写的教科书是人工智能领域最被盗版的教科书，他肯定会告诉你。他非常担心自主武器，以及更广泛地说，如何构建我们仍然可以控制的AI，换句话说，不会超越我们并想做对人类不利的事情。所以我对这次对话非常感兴趣，但在我的旅途中，我遇到了你们，我很想了解一下。
So Lister is kind of understand kind of what primer is, how long you've been around, and kind of how you got going. And then we can talk about all the interesting slash scary things that are happening in the world right now.
所以，Lister有点理解Primer是什么，你已经存在了多久，以及你是如何开始的。然后，我们可以谈论当前世界上发生的所有有趣/可怕的事情。
Yeah, look absolutely. So I started primer in 2015. Prior to that, I'd come out of computational physics background. And spent a lot of time modeling insurgents, these in places like Iraq and Afghanistan and building computational models.. To try and explain how insurgents would organize themselves.
是的，完全没错。所以，我在2015年开始进行底漆。在那之前，我从事计算物理背景，花了很多时间模拟叛乱分子，这些地方像伊拉克和阿富汗，并建立计算模型。以尝试解释叛乱分子如何组织起来。
And effectively why they were so hard to defeat. And so that was, you could do that with physics?
实际上，为什么他们很难被打败？那么，你能用物理学做到这点吗？
Well, so we were sort of the ugly child of physics until we published the work on the cover of nature. And then we were their favorite. When you say we, who was we? Yeah, so myself, my supervisor, and there was a couple of other people on that. There's about I think five authors on that paper.
嗯，所以我们在发表了《自然》封面文章之前，一直都是物理学的丑孩子。然后，我们就成了他们的宠儿。你说的“我们”，指的是谁呢？是的，我、我的导师和还有几个人都参与了这项研究。那篇论文大约有五个作者。
And I think it was the first time nature that ever published an analysis of conflict or quantitative analysis of how insurgents work. And it was really, you know, the first sort of reaction to that stuff was like, go and put them in the political science space.
我想这是自然杂志第一次发表关于冲突分析或反叛分子工作的数量分析。当时的第一反应实际上是将它们放在政治科学领域中。
Yeah, yeah, yeah. When was this? Where you? It published in 2008. Okay. So it was kind of like right at the cusp of the sort of, you know, the big data revolution was like actually.
嗯，嗯，嗯。这是什么时候？你在哪里？它在2008年发表的。好的。所以它就像是那个大数据革命的转折点一样。
A new student. Yeah, I was a PhD student. Yeah. Yeah. Yeah. Got you.
一个新学生。是的，我是一个博士生。嗯。嗯。嗯。我明白了。
So I had a lot of kind of interest in physics systems to try and explain the world that we're living in. Right. And as part of that, this sort of like brought into a couple of places. One was we were using unclassified information around attacks, where they happen, when they happen, and so forth. So we got into really primitive in LP stuff back in 2008.
所以我对物理系统有很多兴趣，试图解释我们生活的世界。对。作为其中的一部分，这引入了几个地方。其中之一是我们在使用未分类信息来了解攻击时发生的情况以及发生的时间等方面。因此，我们在2008年开始使用非常原始的LP技术。
And the second bet natural language process, natural language processing. Yeah. Which has just come so far like to what we are today. I couldn't have imagined how far it would have come.
第二个是自然语言处理，就是自然语言处理。嗯。它已经发展到了我们今天的程度。我无法想象它已经发展到了这个程度。
And the second bet was agent based modeling and using reinforcement learning to start to kind of model insurgent dynamics to try and match the information that we're picking up.
第二个赌注是基于代理的建模和使用强化学习来开始模拟叛乱动态，以尝试匹配我们收集到的信息。我们想要通过这种方法来更好地理解叛乱活动。
Really trying to get a handle on, you know, how insurgents work and why they're successful. That's that was my PhD work. And so you created that and that kind of put you on the map, so to speak.
我真的很想了解游击队是如何运作并且为什么会成功，这就是我的博士研究的内容。因此，你创造了这个，并且这让你声名鹊起。
And I was back in 2008. So how did you end up starting primer and kind of what was the path there?
那时我回到了2008年，你是怎么开始创办Primer的？整个过程是怎样的呢？
I mean, so after that, I mean, there was interesting. So now you kind of get meshed into the world of counterinsurgency. And I did that probably for a couple of years. I spent as an advisor, I spent, you know, a couple of months in Northern Iraq with the deputy prime minister of Iraq.
我的意思是，之后，我是说，那是有趣的。现在你有点被卷入了反叛世界。我做了这个工作大概两年。作为一名顾问，我和伊拉克副总理一起在北部伊拉克待了几个月。
And that was where? Urbiel. I've been an air bill. You've been her bill? Yeah. A few times. So I used to cover oil. Yeah. So I did a bunch of stories on kind of Iraqi Kurdistan and their attempts to kind of create the oil industry and all of that stuff. And yeah, I mean, it's her year of bill.
那是在哪里的？乌尔比尔。我曾作为一名记者到过那里。你曾为她打工？是的，有几次。我曾报道伊拉克库尔德斯坦的石油工业和相关的新闻。 是的，这是她的一年。
I think I was there. I remember watching the election results come in. It was the Obama results with with the deputy prime minister of Iraq and just being like, this is this is a strange thing. But yes, it took me some interesting places.
我记得当时在那儿。我看着选举结果出来。那是 Obama 与伊拉克副总理的结果，我只是感到这是件奇怪的事。但是，这使我去了一些有趣的地方。
It took me to presentations the UN that took me briefing the young officers that were about to leave to go from West Point. And it was kind of like everyone at the time was kind of struggling to say, well, how do we really understand this opponent? And you know, if you look at it, you had an insurgency was taking on and defeating the strongest military in the world. Perhaps the strongest military the world's ever seen. And so we didn't certainly have all the answers, but you know, we had some. And so that took me around all of that.
我曾在联合国做了两次演讲，向即将离开西点军校的年轻军官们进行了简报。当时每个人都在努力想，怎样才能真正了解这个对手？如果你看一下，你会发现他们在与最强大的世界军事力量抗衡并战胜了对方。也许是世界上最强大的一支军队。我们并没有全部的答案，但我们了解了一些情况。因此，这件事带着我走遍了这一切。
But I ultimately came back and I was like, if I'm going to do anything here, it can't just be writing scientific papers. The theories need to kind of be put into actions with tools. And you know, it was 2009 and I was like, I don't know everything, but I know one thing that is like if you come anywhere in the world, you build in Silicon Valley.
但我最终回来了，并且我觉得，如果我要在这里做点事情，不能只是写科学论文，理论需要用工具把它们实现。当时是2009年，我不知道所有的事情，但我知道一件事情，那就是如果你来到这个世界的任何地方，你都要在硅谷建立。
And so I came out here. I think, you know, maybe had, you know, my two suitcases and five thousand dollars in my pocket and slept on friends couches. And I was just like, I just need to be here. And I've got a sense that I need to kind of build. And so that sort of started that journey off and started my first company company called Quid. And we were visualizing high dimensional data structures and it was super interesting and learned all sorts of stuff and made all sorts of mistakes.
所以我就来到这里了。你知道的，也许只带了两个行李箱和五千美元在口袋里，睡在朋友的沙发上。我就觉得我需要在这里。我有一种要建立的感觉。于是这开始了我的旅程，创办了我的第一家公司Quid。我们可视化高维数据结构，这很有趣，学到了很多东西，也犯了很多错误。
And then we we sold that company and it was nice, but 2015 came around. And I'd seen my friends building these big computer gaming rigs and training these large kind of like image recognition models. And the results were just incredible. And I was like, this changes everything.
然后我们把那家公司卖掉了，感觉不错，但到了2015年。我看到我的朋友们正在建造这些大型电脑游戏设备，训练这些大型的图像识别模型。结果简直是让人难以置信。我想，这改变了一切。
What made you think that? Well, you just so the benchmarks that we were saying in image recognition problems were jumping like 30 points. So image recognition was was in, you know, if you go back to sort of 2010 or just when I was finishing PhD work, my friends at the computer science department was like, this, this is impossible. This is this is something that machines can't do is what humans do. And so you see machines doing it.
你为什么会这么想呢？嗯，你看我们在图像识别问题中提到的基准标准跃升了大约30分。所以，图像识别，你知道，如果回顾一下2010年左右，我刚完成博士论文的时候，我的计算机科学系的朋友们都说，这是不可能的。这是我们人类才能做到的事情。但现在你看到机器做到了。
And I remember just shaking my head and trying to like. Like put images in front of this computer. I was like, it was guessing them. And I was like, right, I need to learn about this. So this is deep learning. And I'm like, all right, this changes everything. And so I thought about that. I was like, two things. One was like, it's probably going to have a big impact on language. Not that the images is as important, but most of us work with language. And the second bit was, I want to get back into the defense intelligence problem space.
我记得当时只是摇头，然后试着把图像放在这台电脑前。我大概是在猜测它们。我想，好的，我需要学习一下深度学习。我就这样想着，这改变了一切。所以我考虑了一下，有两件事，一是它可能会对语言产生大的影响。虽然图像也很重要，但我们大多数人都使用语言。第二个是，我想重新进入国防情报领域。
I've been out of that for about five or six years. And it was like, it's time to get back into the defense space. And so what was quid? So quid was focused on data visualization. So you take these data sets, which could be anything from sort of comments on products through to advertising, copy through to anything like big data. You'd have these the all this text based information.
我已经离开那个领域大约五六年了。当时我感到，是时候重新回到防御领域了。于是Quid应运而生。Quid专注于数据可视化。你可以处理不同类型的数据集，例如产品评论、广告文本或大数据等等。你会处理所有这些以文本形式呈现的信息。
And you're like, well, what do I do with 5,000 comments about this product? Right. And what we said was actually you can visualize that navigated and start to see all the different clusters of things that emerge in a visual way. So you can start to get a handle on the narratives that are unfolding. And so I was really kind of saying, look, natural language processing at that time was not as advanced as it needed to be, but we can bridge the gap between where the technology is. And where value is by visualizing things.
你可能会想，我得怎么处理这五千条关于这个产品的评论呢？没错。我们实际上说的是你可以将其可视化并开始以视觉方式看到不同的聚类现象。这样你就能够开始掌握正在展开的故事情节了。所以我实际上是在说，那时的自然语言处理技术还不够先进，但我们可以通过可视化来填补技术和价值之间的差距。
Right. Right. So it was using some really, really interesting visualization techniques to kind of allow people to interact with at the time what we're cutting edge natural language models. But today would be seen as being very primitive. Yeah. Right. So you started primary work in 2015.
好的，那它使用了一些非常有趣的可视化技术，以便让人们与当时的先进自然语言模型进行交互。但是今天看起来已经非常原始了。是的，你从2015年开始了主要工作。
Yeah. Yeah. What was the idea? So the idea was basically to come in and say, look, there's two macro trends. One is like defense is going to be a bigger, bigger issue. And the second is we're going to we should expect rapid performance improvement in language understanding because what we were saying from deep learning and images. So those are the two things is, you know, we don't know exactly how that's all going to map out, but that's exactly where you want to be.
嗯，嗯。那这个主意是什么？基本上的主意是，我们要来说，看，有两个宏观趋势。一个是防御会成为一个越来越重要的问题。第二个是，我们应该期望在语言理解方面快速提高性能，因为我们从深度学习和图像中所说的。所以这两件事情是，你知道的，我们不知道它们会如何具体实施，但这正是你想要的。
Right. And so what that meant was firstly was assembling a very strong technical team to go after those problems. And then secondly, it was looking as saying, right, we think the first place is going to land as in the intelligence community. Right. They deal with text. They're getting more and more text. And this is going to be the place we land. So they became the first customers.
"好的，那就意味着首先我们要组建一个非常强大的技术团队去解决这些问题。其次，我们要看一下，我们认为第一个需要解决的地方是情报界。他们处理文本，而且获取的文本越来越多。所以这是我们要进军的领域。因此，他们成为了我们的第一批客户。"
And we took financing really early on from in Qtel, which was the investment arm of. They say the intelligence community, but yes, you can you can infer kind of maybe who might be behind that. So we took money from in Qtel and got working on automating a number of processes that intelligence analysts do want on a regular basis. And one of those is building knowledge graphs.
我们很早就从Qtel的投资部门获得了融资，他们被称为情报界，但是，你可以推断出可能是谁在背后。因此，我们从Qtel那里拿到了钱，并开始自动化许多情报分析师定期需要做的流程。其中之一就是建立知识图谱。
Right. So there's a quite a manual process. This person is the same as that person, which goes by this alias. They travel to this location. They met this person. They talked about these things. So you've got a kind of a graph, right. And you can kind of fill it out manually, but it's very consuming right now without going to the classified.
好的，所以这是一个相当冗长的过程。那个人是同一个人，使用这个化名。他们去到这个地方，遇到了这个人，谈论了这些事情。所以你拥有了一种图形，对吧。你可以手动填写，但现在需要去查阅分类信息，这非常费时间。
So we know from things like Wikipedia, the recall of everyone that has Wikipedia pages that should have Wikipedia pages or put another way of everyone who should have a Wikipedia page. How many actually do is depending on the metrics about a third, right. And it skews obviously along certain traditional biases.
所以我们从像维基百科这样的事物中知道，所有应该有维基百科页面的人都被召回，或者换句话说，所有应该有维基百科页面的人。实际拥有多少页面取决于度量标准，大约有三分之一的人拥有页面。这显然会偏向某些传统偏见。
So even with things like Wikipedia, where you have a huge kind of crowdsourced dynamic, we don't we do a terrible job of keeping that information up to date and accurate and accurate and then it's not just do they have a page, but how long from when information changes does it get updated. Right. And the latency on the stuff can be about six months. Yeah. Right.
即使有像维基百科这样的东西，我们也很难做到将信息保持最新和准确无误，因为它是由大量人参与的动态过程。不仅如此，信息更新的时间也很关键，不只是页面的存在与否，而是我们需要多久来更新它。就像这样，这种信息的延迟可能长达六个月。
So that's one child right building and maintaining knowledge graphs. And that's one that intelligence has in spades. What does that mean? It means like one thing they talk about is you can have a request to say look, we've just had a pro US demonstration at Iran. What should we expect the actions of the Iranian Revolutionary Guard to be in the next month.
那么这就是一个孩子，专门建立和维护知识图谱的。而情报机构在这方面可谓是造诣深厚。这是什么意思呢？这意味着，他们谈论的其中一件事是，你可以提出这样的请求：我们刚刚在伊朗看到了亲美示威活动，接下来一个月内伊朗革命卫队会采取什么行动呢？
And you say, well, I don't know like I could give you thing or you could say, well, I don't know what have they done every time there was a pro US demonstration in the past and we say, all right.
然后你会说，我不知道，我能给你什么东西。或者你也可以说，每次以前有支持美国的示威活动时，他们都做了些什么，我不知道。我们说好。
Now traditionally to solve that, you now have like a dozen intelligence analysts working for three weeks to kind of put down all that information, write it up and then do an analysis with a system like hours you just go through and say, give me all the events that followed immediately after a program.
传统上，要解决这个问题，你可能需要十来个情报分析员连续工作三周，整理所有的信息并撰写报告，然后再用类似小时这样的系统进行分析，并查看所有在节目之后立即发生的事件。
And then we after a pro US demonstration within 50 miles of Iran and then categorize them according to the ontology that I care about. And so now you've got that data in the space of minutes with one person at the wheel.
然后我们在距离伊朗50英里内参加了一次亲美示威活动，然后根据我关心的本体论对它们进行分类。因此，现在你只需要一个人操作就可以在几分钟内获取这些数据。
You know, conversely like if you had to find all the locations that someone's been or travel to again, I can go through and read every document and figure out where they traveled or I can just say person A has traveled to location.
你知道的，从相反的角度来看，就好像你不得不找到一个人去过的所有地方，或者再次旅行，我可以查看每个文件并确定他们去过哪些地方，或者我只需说A人已经去了某个地方。
So what you've built, kind of like an LLM or like a large language model, but for intelligence.
你所构建的东西，有点像LLM（大规模语言模型），但用于智能。
Yes. So what we've seen since 2015 really has just been a continual but pretty rapid increase in capabilities of language or language models. And so what that means, concretely for us and our customers is yes, we deploy large language models into intelligence environments to allow them to interact with the data in a way that firstly reduces the time.
是的。自2015年以来，我们看到的是语言或语言模型能力的持续但相当快速的增长。因此，这就具体地意味着对于我们和我们的客户来说，我们将大型语言模型部署到智能环境中，使它们能够以一种减少时间的方式与数据进行交互。
But what we really say is it allows you to be more curious and for an intelligence analyst, if I tell you, you know, hey, I've got a hunch that there might be a patent from the Iranian Revolutionary Guard response. Give me three weeks to go and think about it. That's very different from I've got a hunch it will take me 30 seconds and I'll see if that hunch is correct. And as an intelligence analyst, it's not that you don't have the data available.
但是我们真正想说的是，这使你变得更加好奇了。对于一个情报分析员来说，如果我告诉你，嘿，我有一种感觉伊朗革命卫队可能有一个专利应答，给我三个星期去思考一下。这和我有一种感觉只需要30秒钟，看看这个感觉是否正确是很不同的。作为一个情报分析员，你并不是没有可用的数据。
It's that the cost of answering the questions is so high that you don't necessarily connect the dots.
问题是回答的成本太高了，你并不一定能够连接起来。
Given where we are, because it does kind of go back to where we started with chat, GBT kind of Stuart Russell called it. He referred to it as like a wake up call for kind of everybody for the public, but also for governments who are like, oh, the stuff is cross some threshold of usefulness. And now it's going to change a lot of stuff pretty quickly.
从我们现在的情况来看，因为它有点回到了我们开始谈论聊天时的地方，GB 头儿称之为一次警钟。他认为这是一个对公众的醒悟，同时对政府也是这样的。他们意识到这些技术已经越过了某些有用的门槛，现在将非常迅速地改变许多事情。
From where you sit, does this moment or AI broadly, I know AI is kind of this squishy term, does it change war?
从你所在的位置看，这个时刻或广义上的AI，我知道AI是一个比较模糊的说法，它会改变战争吗？
I think it's good to kind of split AI into a few different buckets, right? And the first is AI for autonomy. And that's kind of a combination of different things, but there's AI for autonomy. There's generative AI, which is AI to create things. And then there's what I would kind of characterize as AI for effectively search. I think when we think about AI, you're across those places.
我想把AI分成几个不同的桶可能更好，对吗？第一个是用于自主的AI。这是不同事物的结合，但有自主AI。还有生成AI，即用于创建东西的AI。然后，我会把AI的有效搜索描述为一种AI。我认为当我们思考AI时，我们会涉及这些领域。
Search can be text, but it can also be images, right? And so you're effectively saying to the AI, I look at the world, text images, audio, and structure it in a way that says when I see a car, it's a car, when you refer to this person, is that person, etc.
搜索可以是文本，也可以是图像，对吧？所以你实际上是在告诉人工智能，我看到世界是文本、图像、音频，并将其构建成一种方式，即当我看到一辆汽车时，它是汽车，当你提到这个人时，就是那个人，等等。
So now if we step back on this here and look at how artificial intelligence is going to impact war, I think chat GPT is probably less kind of like relevant than maybe some of the other pieces.
所以现在如果我们回过头来看看人工智能将如何影响战争，我认为聊天GPT可能比其他一些部分更不相关。
I would say first and foremost, autonomy, having systems that can autonomously navigate, move, make decisions, like that's very, very clear. And we know even from the early simulations that I think the writings very much on the wall that humans will not fly single planes better than machines.
首先，自主性非常重要，即拥有可以自主导航、移动、做出决策的系统。从早期模拟中我们已经知道，人类无法比机器更好地驾驶单独的飞机，这一点非常明显。
You referenced this when last we spoke, but most people don't know this because so could you explain what you're talking about?
上次我们交谈时，你提到过这个，但大多数人并不知道这个是什么，请问你能否解释一下你所说的是什么？
So we've all been watching Top Gun, right? Maverick and he's back and he's like, that is pure nostalgia, right? Like the reality here is, so DARPA has been running a series of tests, they did their first head to head competition. DARPA is the Pentagon's kind of tech investment in the research.
我们都在看《壮志凌云》对吧？Maverick回来了，这真是纯粹的怀旧啊。现实情况是，DARPA一直在进行一系列测试，他们进行了首场竞争。DARPA是五角大楼在研究领域的技术投资。
Yeah, for looking big defense bets, like they really ran the first self-driving car test, I think back in 2000 and maybe 7,6. And so the latest one they did, well, one of the latest ones was Canada machine pilot, and I think there were 16s, but I don't know maybe there were 22s.
嗯，对于看起来很有把握的防御赌注，比如他们在2007年或2006年进行了第一次自动驾驶汽车测试，我认为这很了不起。他们最近进行的一个测试是加拿大机器驾驶员，我认为有16个人参加了，但也可能是22个人参加了。
Pilot a plane better than a human Top Gun. And they ran this, there were five, I think five or six teams that entered the winning team that win against the human.
驾驶飞机比《壮志凌云》中的高手驾驶更好。他们进行了比赛，参与了五个，我想是五个或六个团队，其中获胜的团队打败了人类。
So it was a human Top Gun and a simulator against a machine. In a dog fight, in a dog fight, one on one, right? One on one. The machine beat the human 5-0, not even close. And the human came back and said, or some of the commentators maybe said it wasn't fair because the machine took risks that the human wouldn't take. And I'm like, yeah, exactly, right?
所以这场比赛是人类的翘楚与一种模拟器对抗机器。在一场狗斗中，是一对一的，对吗？就是一对一。机器5-0轻松地战胜了人类，根本没什么悬念。然后那个人类回来说，或者一些评论员说机器胜利不公平，因为机器能够冒险而人类不愿意。然后我就想，恰恰就是这样对吧？
They said, you know, the machine was able to kind of like shoot before the human reaction. So it wasn't really a fair fight. And he's like, well, exactly.
他们说，这台机器有点像在人类反应之前射击。所以这并不是一场公平的比赛。他说，确实是这样。
Right? So it's kind of like you have tanks and horses and you're like, well, it's not really fair.. Obviously the tank is going to be faster because it's not a horse. So you don't have to feed a tank and. Right. And so look, you know, you don't need to put a human in these machines. They can take risks. You don't have body counts associated with them. They've got higher G's that they can pull because they don't need to have a human blacking out. Right.
那就像你有坦克和马，你会觉得不太公平。显然，坦克会更快，因为它不是马。所以你不需要喂坦克。对吧。所以你看，你不需要把人放在这些机器里。它们可以冒险，没有相关的伤亡人数。它们可以承受更高的G-力，因为不需要人类昏倒。对吧。
So there's all these advantages. And it's pretty clear the machine is one. Now, they're moving to live in air tests now, some movement from the simulations to in the air, which is simulations of that good these days. I don't think we should expect the results many different. How they're going to do that? I mean, I guess it's just idle pilot on steroids. It's auto pilot on steroids. And artificial intelligence.
所以有很多优点。显然，其中之一就是使用机器。现在，他们正转向进行实际空中测试，从模拟测试向实际空中测试转移，而现在的模拟测试效果已经很好了。我认为我们不应该期望有太大的不同结果。他们打算如何做到这一点？我的意思是，我猜想这只是单纯的“类固醇式”的自动驾驶。加上人工智能。
I mean, one way to think about it is auto pilot on steroids. And it's got so good now that it's better than the best human. And the next step on that. So Maverick and Goose are screwed. Yeah, the Goose is cooked, right? Yeah. That's right. You know, so that's gone. But then it's not going to stop at one plane. It's going to be a swarms of planes. Yeah. Right. And or swarms of drones or swarms of drones.
我的意思是，从一个角度来看，这就像是超级厉害的自动驾驶。现在这技术已经非常成熟了，已经比最厉害的人类驾驶员还要好了。未来的发展方向就是这样。所以，Maverick和Goose就完了。是啊，目标已经被锁定了。这个可以不用管了。但这可不止一架飞机。会有一批飞机组成的群体，或者是无人机。
And all of this is going to move because, of course, you can manufacture these things cheaper, they're more disposable. So you're going to have swarmed the swarm conflicts. And this also brings back to kind of some of the stuff that we studied with our colleagues in physics or swarm dynamics. You know, how do you model biological swarms and all the rest of it? So that I think is going to be really interesting on the autonomy side.
所有这些东西都将会改变，因为显然你可以更便宜地制造这些东西，它们也更易于处理。因此，你将会有一堆堆的卫星冲突。这也会让我们回想起我们和物理学或群体动力学领域同事研究的一些东西。你知道，如何模拟生物群落等等。所以我认为在自治方面这将非常有趣。
Do we know when the actual in air dog fight, man versus machine is happening? You know, I hope this gets as much love and attention as the AlphaGo match or the Watson Janet imaginary will get less. You know, it should because it's kind of crazy, right? Like, you know, we're putting machines for the Battle of S, air supremacy machines will either give it to us or give it to our opponents. But it's not going to be humans doing it. And do we know if that is that happening like this year? You know, I think it is, right? Like I saw the photo shoots of a plane I think that they're going to put the controls in. You know, that came out, like maybe a couple of weeks ago.
我们知道实际的空中犬战，人类对机器的战斗会在什么时候发生吗？你知道的，我希望这场比赛能够得到与AlphaGo比赛或Watson Janet虚构比赛一样多的关注和喜爱。你知道的，这是有点疯狂的，不是吗？就像，你知道的，我们把机器放在S区的空中优势战争中，机器要么给我们，要么给我们的对手。但这不会是人类在做这件事。我们知道这是不是在今年发生的吗？我想是的，对吧？就像我几周前看到的那张可能要放置控制装置的飞机的照片一样。
I need to bone up on my like, you know, air defense weekly, you know, reading because that's quite amazing. That's amazing. So that's going to come down. And I think that should give us another Sputnik moment, or hopefully gives it a different department, another Sputnik moment.
我需要加强我对空防的每周阅读，因为那太神奇了。那真的很神奇。所以，这将会有所下降。我认为这应该给我们带来另一个“卫星一号时刻”，或者希望在不同的部门带来另一个“卫星一号时刻”。
And what that really means, though, is you're OK. Run that forward. You've got a swarm of these things, running by machines, taking on an opponent swarm. And then you say, all right, that's cool. You've got software and AI supporting these. And we think we're pretty good. We've got dominance. But then your opponent upgrades all of the swarm overnight and all of a sudden they have 99% kill rate on your system.
那真正意味着你很好。进一步说，你拥有一群由机器运行的东西去迎战敌对群体。然后你说，那真的很酷。你拥有软件和人工智能来支持这些。我们认为我们做得相当不错。我们占主导地位。但是，你的对手在一夜之间升级了所有的群体，突然间他们系统的击杀率达到了99%。
So once you're into a place where the controlling factor is the AI capabilities that drive the machines, you can push an update like you would with a Tesla. And all of a sudden, your car goes from being OK to being superhuman. Now you're 20,000 drones that are whatever, 1,000 bucks a pop or whatever are just appreciably better than yours. That's exactly right.
当你进入一个控制系统因AI功能而推动机器的场所，你可以像更新特斯拉一样对其进行升级。突然间，你的汽车从普通变成了超级英雄。现在你有20,000个无人机，每个只需一千美元或其他价格，比你的更好。这完全正确。
So you've got the fans you're feeling good about defending Taiwan. You've got all the things. And then overnight, your opponent upgrades the capabilities. You didn't even know the upgraded the capabilities they decide to attack and they beat you. And everything that you've done to that point was rendered useless by an offset in the AI capabilities.
所以你有许多粉丝，你对捍卫台湾感到自豪。你拥有所有东西。然后一夜之间，你的对手升级了能力。你甚至不知道他们已经升级能力并决定进攻，最终打败了你。所有到目前为止你所做的一切都被AI能力的偏移所抵消了。
Now, when it comes to war, we've never seen this kind of speed at which you can achieve an offset across your entire fleet against your opponent. Because over the year update, it's not like you can over the year update the quality of the tanks or your nukes. But you can over the year update the quality of your swarm for your supremacy.
现在，说到战争，我们从未见过这种速度，您可以在整个舰队中与对手实现抵消。因为随着年度更新的进行，您不能像年度更新坦克或核武器的质量那样。但是，您可以通过年度更新来提高您的群集质量，以确保您的霸权。
I'm Alex Dibble from The Times. There's so much happening in the world right now. How do you stay on top of it? On the World in 10 podcasts, near my team, sit down and go through it all in just 10 minutes. The Times Correspondence join us with Eyewitness Accounts and Interviews. I was actually on the orthopedic war, almost all the injuries are shelf-fired. A woman who was inside the police van. They also give us their unique analysis of world events. Give the world in 10 listen. It is 10 minutes to stay on top of the world.
我叫亚历克斯·迪布尔，是《泰晤士报》的记者。现在世界上有很多事情正在发生。您是如何保持关注的呢？在《世界十大播客》中，我和我的团队坐下来花费仅仅10分钟来浏览所有的新闻。《泰晤士报》的通讯员们与我们一同分享目击报告和采访。事实上，我曾在骨科战线上，几乎所有的伤员都是枪伤造成的。还有一名被关在警车里的妇女。他们还为我们提供了关于世界事件的独特分析。请听一听《世界十大播客》。只需10分钟，您即可了解世界最新动态。
In the Uni, I studied political science, which hasn't really done much for me. But I remember studying a lot this idea of mutual issue destruction, which was what kept the cold war cold. Because we had, however many thousands of nukes pointed at them and likewise, and everybody was like, well, once you pressed a button once, then everybody's dead..
在大学里我学习了政治学，但实际上对我并没有什么帮助。不过，我还记得我们学习了相互问题摧毁的理念，这就是让冷战保持冷静的原因。因为我们对着敌人有成千上万个核弹头，反之亦然，每个人都知道，一旦有人按下一次按钮，那么每个人都将死亡。
And it feels like the competition here, and obviously everybody's been talking about this for a while, especially folks like Eric Schmidt, the former Google CEO, the competition with China, this dynamic feels different than that. And I don't know how you think about that, or if that is, are people thinking about that yet? Because it feels, again, going back to like the chat GP team, just a sputnik moment, or when the machine beats Maverick, et cetera.
感觉在这里的竞争，显然每个人都在谈论很久了，尤其像前谷歌CEO艾瑞克·斯密特（Eric Schmidt）这样的人，与中国的竞争，这种动态感觉不同。我不知道你怎么看待这个问题，或者有没有人已经开始思考这个问题？因为这种感觉，再次回到像聊天GP团队，就像苏联的卫星一样，或者当机器打败了女扮男装的主角等等。
It feels like this stuff isn't far away, this idea of being able to do the over-the-air update, and all of a sudden, our machines are just the best in the world, our leapfrog, and then back and forth and back and forth.
感觉这种能够进行空中升级的想法并不遥远，一下子，我们的机器就成为了世界上最好的，我们跨越了同行，来回反复。
But these aren't like a bunch of A-bombs that we just have there pointed. It feels like more back and forth, but I don't know. I'm just trying to kind of create a meta model for what that dynamic is. And it feels somehow very similar, but also very different.
但这些不像一堆只是呆在那里指着的原子弹。感觉更像是来回的交替，但我不确定。我只是试图为那种动态创建一个元模型。感觉有些相似，但也有很大不同。
Yeah, so when it take back, I think, I was back 2016. So 2016, the Office for Net Assessment, which goes back to the Cold War, which is an interesting kind of group from Pentagon, assembles a bunch of AI people, and brings us up to West Point. And I'm there, Stuart Russell is there, there's some hedge fund-quant people there. It's a really interesting group, 15 of us.
嗯，当它被带回来时，我想我回到了2016年。2016年，针对净评估的办公室，可以追溯到冷战时期，是五角大楼一个有趣的小组。他们聚集了一些人工智能方面的人，把我们带到了西点军校。我在那里，Stuart Russell也在那里，还有一些对冲基金量化人士。这是一个非常有趣的小组，一共有15个人。
And we spend the entire week in back-to-back kind of presentations from us and debates and discussions around what impact AI is going to have. This isn't 2016. 2016. Oh, interesting. It was. So we go up there, and it's just they won't let us out of the room.
我们整整一周的时间都在不断地进行演讲、辩论和讨论，研究人工智能（AI）将会有什么影响。这可不是2016年了。2016年？哦，有趣。那么我们就上去了，但是他们就是不让我们出房间。
And it was just going through all night on back and forth. And I was there with Stuart Russell, where they were, it was fascinating. And a couple of things emerge. One is this should be considered as fundamental change to war as the internal combustion engine. The second is it will touch everything. Yep.
那是一整夜地反复讨论。我和 Stuart Russell 一起在那里，那真是很有意思。有几个要点。第一，这应该被视为战争基本变革，就像内燃机一样。第二，这将触及到一切。没错。
And the third piece is this is probably best understood in the concept of offset. And so we presented the work to the Deputy Secretary of Defense Robert Work, who came in, and we briefed him on at the end of the week and so on. And sort of started to kind of formulate around this hypothesis of like artificial intelligence being the third offset.
第三个部分最好用抵消的概念来理解。于是我们向国防部长助理罗伯特·沃克展示了我们的工作，并在周末结束时向他进行了简报。我们开始尝试构建一个假设，即人工智能是第三个抵消因素。
Yeah, so offsets. Yeah. So an offset is a technological advantage, so great that it renders those without that capability defeated before the battle even starts. Right. So to run it back, first offset nuclear weapons. If you have nuclear weapons and your opponent doesn't, no point fighting because I'll drop a nuclear weapon and your entire country's gone. Which is gone.
是的，就是战略偏移。嗯，战略偏移是一种技术上的优势，非常强大，让那些没有这种能力的人在战斗开始之前就被打败了。对吧，所以先是核武器的战略偏移。如果你拥有核武器而你的对手没有，那就没必要打了，因为我可以投掉一枚核武器，你整个国家就没了。就是消失了。
Second one comes through. So it's interesting. So you talk about loot to the arms race of nuclear weapons. One of the things that actually broke that was precision guided munitions. And precision guided munitions basically said, yeah, we're kind of equal in the number of nuclear weapons that we've got.
第二个到了。这很有趣。所以你说掠夺就像核武器军备竞赛。事实上，其中一个打破这一情况的因素是精确制导武器。精确制导武器基本上表示，是的，我们在拥有核武器数量上具有相等性。
But if I can't guarantee my nuclear weapon hits your nuclear weapon, then I need three or four of them. If I can guarantee it, I only need one of them. And if you don't have that precision munition, now I've got a five to one six to one advantage. The reality is it actually moves a lot further. There's a story from Vietnam where they tried to bomb a bridge.
如果我不能保证我的核武器击中你的核武器，那么我就需要三到四个。如果我能保证它，那么我只需要一个。而如果你没有那种精确的武器，现在我有五比一或者六比一的优势。现实情况是它实际上走得更远。有一个来自越南的故事，他们试图轰炸一座桥。
They counted 800 craters. And they couldn't count the ones that were in the water because they couldn't see them. But there were 800 craters around the bridge and the bridge was not hit. Right. And this isn't the 70s, right?
他们数了800个弹坑。但是他们看不见水中的弹坑，所以没法数。但是桥周围有800个弹坑，但桥没有被击中。是的。这可不是70年代了，对吧？
And then they finally got the first semi-conductor powered precision munition that got the bridge. And so what you see as the second offset was precision munitions and stealth weaponry. And we sort of forget that a little bit. But if you look at the first Gulf War, you have the strongest army in the world, which is America, against the sixth strongest army in the world, which is Iraq, and the entire war is over in 72 else. Yep.
然后他们终于得到了第一种半导体动力的精密武器，成功摧毁了桥梁。所以你可以看到第二次变革是精密武器和隐形武器。我们有些忘记了这一点。但是如果你看看第一次海湾战争，你会发现最强大的美国军队对抗世界第六的伊拉克军队，整个战争只持续了72小时。没错。
And so that's the second offset. And then the third offset then becomes artificial intelligence. And artificial intelligence says, if your machines have the ability to beat every human in the air, then the war is over before it starts. Conversely, if your AI can beat my AI, the war is over before it starts.
那么这就是第二个偏移。然后第三个偏移就成为了人工智能。人工智能说，如果你的机器有能力在空中击败每一个人类，那么战争就在它开始之前就结束了。相反，如果你的人工智能能够击败我的人工智能，那么战争也在开始之前就结束了。
But what's different about this is that you can get an offset in AI capabilities that renders your opponent useless overnight. And the thing that should be the wake up call with chat GPT is we had large language models, hundreds and billions of parameters.
No one really thought to put reinforcement learning human feedback into it at scale. It was done. And you went from a reasonably good autocomplete to I can pass an MBA test overnight. Like take that into the defense environment. You have a reasonably good swarm.
这里的不同之处在于，你可以在人工智能方面取得偏差，使你的对手在一夜之间变得毫无用处。而聊天GPT应该警醒我们的是，我们有大型语言模型，数百亿个参数。没有人真正想到要在规模上将强化学习的人类反馈应用于其中。它已经完成了。你可以从一个相当不错的自动补全到在一夜之间通过MBA考试。把它带入防御环境中。你有一个相当不错的群体。
To I have one that is so much better than anything you've ever seen overnight. And that's the bet that the US Defense Department hasn't orientated itself around as that we've seen offsets before. But we've never seen an ability to have an offset overnight. And what that means is as soon as you have that capability, you attack because you know that your opponent is behind.
我有一个比你见过的任何东西都要好很多的东西。美国国防部没有像我们之前看到的那样朝这个方向调整自己的策略。但我们从未见过能够在一夜之间拥有对抗优势的能力。这意味着一旦你有了这种能力，你就会进攻，因为你知道你的对手已经落后了。
And so this arms race ratchets up that you can never let your opponent get more than an offset ahead of you with AI, least they find out that they've got it, and then it's game over. So the speed of this just accelerated massively.
这场军备竞赛不断升级，你不能让对手在人工智能方面领先你太多，否则他们会发现自己已经掌握了这种技术，那么这场比赛就结束了。因此，这种竞赛的速度大大加快了。
This is a clumsy analogy. But it's like the app storeification of defense where it's just like, oh, I've got TikTok on my phone. I just did a story this week on TikTok. And you know what I keep about this new filter.
这个类比有点拙劣。但是就像国防变成应用商店一样，就像说我的手机上有TikTok，我这个星期还做了一个关于TikTok的报道，你知道我对这个新过滤器感到满意。
That turns very normal looking people into like to look like models. Terrible for kids. But I couldn't access it. They're like, updated. And you get this whole new array of filters. And it's just like, but like that with AI and the consequences are mass death or supremacy or whatever.
这会让普通的人看起来像模特一样，真可怕，特别对孩子们不好。但我打不开。它们更新了。你会得到一整套新的过滤器。就像那样，但是这次是用人工智能，后果可能是大规模死亡或至上主义之类的。
Right, because you've got the offset. And once you're able to get that offset over your opponent, you jump through that window. And this is where the thing comes right. It really matters in AI how fast you get that into the hands of the war fighter, how fast you get into the hands where it can make a difference.
对的，因为你有偏移量。一旦你能够把偏移量置于对手之上，你就跳过那扇窗户。这就是事情变得正确的地方。在人工智能中，重要的是你能够把它快速地交到战士手中，让它产生作用的速度有多快。
And the issue here, people talk a lot about an AI arms race with China. They talk about chips. They talk about training data. They talk about research and models. The long poll in the tent is how long it takes to go from something that could be done to something that is being used.
在这里的问题是，人们经常谈论与中国的AI军备竞赛。他们谈论芯片、训练数据、研究和模型。关键在于，从能够做到某事到实际使用需要多长时间。
Right. And what we've seen here is that the US and the procurement cycle has been set up in a Cold War mentality, which says. I'm going to take 15 years to get my F-35. Right. We're number one. So we can operate on our own timeline. And secondly, most of the things we're buying, a large cap X expenditure piece is a metal.
好的。我们看到美国的采购周期是在冷战思维下建立的，这意味着需要15年才能获得F-35。对的，我们是第一。所以我们可以按照自己的时间表操作。第二，我们购买的大多数东西，特别是大型支出的金属零件。
Versus, we've got competitors that are ahead of us. We don't control the timeline. And things can be upgraded overnight that render everything you've done previously useless. And so when you're in that mindset, which is hardware, powered by software, you need to be able to procure technology on the time frame of potentially hours of an app update. An app update. So you look through those capabilities.
与我们相比，在竞争对手方面，他们领先我们。我们无法控制时间表。而且，一夜之间可能会进行升级，使之前所做的一切都变得无用。因此，当你处于这种硬件由软件驱动的思维状态时，你需要能够在可能的几个小时内采购技术，以应对应用程序更新。应用程序更新。因此，你需要浏览这些能力。
And this is the thing that I think is the single biggest danger that the US is facing is our inability to get the technology that we can all produce here in Silicon Valley into the hands of the people that can actually make a difference with it faster than our opponents in China. And China is a long way ahead of us in that military civil fusion than we are.
我认为美国面临的最大危险就是我们无法让硅谷创造的技术更快地交到那些能够用它们去改变世界的人手中，而中国在军民融合方面已经比我们领先很远了。这是我们需要警惕的地方。
What does that look like on the ground? In other words, because when we chatted the other week, it feels like America is ahead in certain aspects. China is ahead in certain aspects. What does that competitive landscape look like? And what does it mean for?
这个问题在实际应用中是怎样的呢？换句话说，因为我们上周聊天时感觉像美国在某些方面领先，而中国在某些方面领先。那么这个竞争格局是什么样子的呢？对于我们来说这意味着什么？
Because I feel like the other thing is, which is a theme, talking to all these people in different parts of the AI world, is that some people are surprised by all of a sudden AI and everything. Some people are like, well, this has been the direction of travel for the past 10 years.
因为我感觉另一个问题是，也就是主题，就是和AI领域的不同人进行交流时，一些人会对AI和一切的突然出现感到惊讶。而一些人则认为这是过去10年的发展方向。
Since that 2012 deep learning moment, image net moment, where I was like, oh, these things actually work really well. This is a nude paradigm. I'm just trying to like, game out kind of, everybody's been talking about this arms race, AI arms race, China for years. But it just feels like, OK, sure. I don't even know what that means. Not quite who cares, but kind of who cares?
自2012年深度学习时刻，图像网络时刻以来，当我意识到这些东西实际上能很好地工作时，这是一个全新的范式。我只是试图推测，每个人都在谈论AI军备竞赛，中国已经几年了。但感觉是，好吧，行。我甚至不知道那意味着什么。不完全是谁在意，但也有点谁在意？
Like, OK, yeah, sure. The thing is about an AI arms race is it doesn't really have an impact until you go to war. Totally. Right? And when you go to war, it really matters. So I was thinking with the deputy minister of technology from Ukraine.
“就像，对呀，当然。关于人工智能军备竞赛的事情是，在你真的参与战争之前，其实并没有什么实际影响。但是，对吧？当你陷入战争的时候，它就真的很重要了。所以，我跟乌克兰的技术副部长商量了一下。”
So he came across to DC and spent some time with him. And you know, what's top of mind for him was how do you label data faster for image recognition for the drones that they've got? That said, what do you mean?
他来到华盛顿和他一起度过了一些时间。你知道的，他最关心的是如何更快地为他们得到的无人机进行图像识别的数据标注。也就是说，你的意思是什么？
Because they will look, we've got things that are spotting objects in the field, but then the Russians will upgrade that. And then we've got to go back and label thousands and more things and, you know, away it goes. If we can shorten that down, we have an advantage. Right?
因为他们会观察，所以我们需要在领域中发现物体的工具，但俄国人将对此进行升级。然后我们要回来对数千个以上的物体进行标记，你懂的，这个过程会不断重复。如果我们可以缩短这个过程，我们就有优势了，对吧？
And so you look at this thing and when you're in war, the inches matter. Because it's a game where it's obviously adversarial and it's competitive in those inches matter.
所以你看着这个东西，当你在战争中时，每一英寸都很重要。因为这是一个明显具有对抗性和竞争性的游戏，英寸很重要。
Now, as we look at this, I think we, firstly as Americans, and most of the West, I think we think conflict with China is a long way off, right? It could never happen. We also thought 12 months ago a land war in Europe could never happen. And that's changed obviously, you know, and changed our reality.
现在，当我们看到这个问题时，我认为我们，首先作为美国人和大多数西方人，认为与中国的冲突还很遥远，对吧？可能永远不会发生。一年前，我们也认为欧洲的陆地战争永远不会发生。但我们的想法显然已经改变了，这也改变了我们的现实。
If you look at the dynamics of the statements from the Admiral Guil Day and the other, you know, folks that are orientated around decisions, they'll say, we have to be ready to fight tonight, right? What that means is that any day that we were sitting here, we may get the call up to send naval vessels into fight in the South China Sea.
如果你看看阮吉尔一天将军和其他以决策为导向的人所说话语的动态，他们会说，我们必须随时准备好进行战斗。那就意味着，任何一天我们都可能收到命令，要派出海军舰船在南中国海进行战斗。
If you look at the sort of the predominant view in Washington DC, it's, there's a five year window with which we expect China to invade Taiwan. And that's credible.
如果你看看华盛顿特区的主流观点，他们认为中国有一个五年的时间窗口会入侵台湾。这个说法是可信的。
Because I, you also know people in like the military industrial complex, they're kind of in a way talking their book, you know, because it's like, oh yeah, of course. So buy a bunch of our stuff. Guil Day is not buy a bunch of our stuff. He controls the fleet. Yeah. The people that are sitting there on the National Security Council, they're not buy our stuff. They control the decisions.
因为我和你一样认识像军工复合体这样的人，他们有点在为自己的利益说话，你懂的，就是他们会说：“噢，当然买我们的东西啦。”但吉尔·戴不这样，他控制着这支舰队。是的，那些在国家安全委员会的人也不是购买我们的产品，他们掌握着决策权。
The dynamics of this is being driven as much from the US Department response to a perceived threat coming in from China. So that's one side of it.
这种动态是因美国政府对来自中国的视为威胁的反应所驱动的，这是其中的一个方面。
The second side of it there that comes through is even if you don't have intention to go to war, accidents can ratchet things up significantly. And as you look at that, I think, look, Taiwan is preparing for an invasion. Certainly you can see China is making strong steps towards that.
这里其实有第二个问题，即使你没有去打仗的意图，意外事件也会使局势急剧升级。所以我认为，看着台湾正在为一场入侵做准备，而中国显然也在朝着这个方向迈出坚实的步伐。
And I think, you know, if I'm putting on my hat and saying, look, if I'm G, an armchair quarterbacking, I'm like, yeah, this is something I want to take back and be part of China. My reign as emperor is not complete without a reunification of Taiwan back into China.
我觉得，你知道，如果我戴上帽子并说，看，如果我是G，就像一个只坐着发表意见的人，我会说，是的，这是我想要取回并成为中国的一部分的事情。如果我作为皇帝的统治不能完整，没有台湾重新归于中国。
And I think looking at that dynamic of all the things here is like, it's, I think, much more likely than not that within a five year window that there is an attempt to take Taiwan back by China. And the US will be, I believe, forced into action. And that's a very difficult set of decisions to make around that.
我认为看这里所有事情的动态，就像是，在未来五年内，中国会试图收回台湾，这是非常有可能的。我相信美国将被迫采取行动。这是一系列非常困难的决定。
But ideally that doesn't happen. And I think one of the things that stops that from happening is having an offset, which is if you were to do this, you'll surely lose, so don't even bother.
理想情况下不会发生这种情况。我认为防止这种情况发生的一个方法是采取抵消措施，如果你这样做，你肯定会输，所以不要浪费时间去尝试。
Not quite mutually assured destruction, but like, look at it. Look at it. We've got the bigger stick. Yeah, you're sure destruction the CCP, right?
不完全是相互保证毁灭，但是，看看这个。看看这个。我们有更大的棍子。对吧，你们确信能毁灭中共，对吧？
And if we maintain an offset in AI capabilities that can be credibly posed back against the CCP or the people's Republican army, then they're not going to attack. If they believe that it's a chance or that they've got the jump, then they'll take that window.
如果我们能够在人工智能能力上保持一定的差距，可以作为回应中共或人民解放军的可信能力，那么他们就不会发动攻击。如果他们认为有机会或者他们已经占了优势，那么他们就会抓住这个机会。
One thing that occurs to me is that, which is a fact that I think a lot of people don't appreciate about Silicon Valley is that from the very beginning, it was kind of hand in glove with the Pentagon and the defense industry. And a lot of those early bets were funded by Washington, you know, kind of going back decades and decades.
有一件事让我想到，我认为很多人没有意识到硅谷的一个事实，那就是从一开始，它就与五角大楼和国防工业手牵手，许多早期的投资都是由华盛顿资助的，可以追溯到几十年前。
And then we've had this turn in the past 10, 15 years. I've heard about it. Others have of just like, you know, folks at Google writing letters to this leadership and like, we do not want to make war machines. We don't want to help the Pentagon with their, you know, project may have been drone, contract, et cetera.
过去的10到15年里我们经历了一个转变。我听说过这个转变，别人也听说过，就像谷歌的员工写信给领导，表示他们不想制造战争机器。他们不想帮助五角大楼完成无人机或其他合同项目。
From where you sit, what does that relationship like today? Because like I said, it does feel like it was very hand in glove. Like it was kind of, there would be no Silicon Valley without the Pentagon.
从你的角度来看，今天那个关系是怎么样的？因为就像我说的那样，感觉像是非常密不可分的关系，就好像没有五角大楼就没有硅谷。
There seem to be have been a breach and without kind of judging whether that's good or bad, but just that as a fact, where are we right now?
貌似存在着一些漏洞，而不去判断这是好事还是坏事，只是作为事实，我们现在处于什么地步？
I think it's bad. Oh, I'll judge it. It's bad. I mean, like, it's like, it's like, you know, you're crying. It's like, you know, imagine Ukraine is just like, oh, like we don't think we should have war. We don't think we should contribute to like any, any like defense stuff. And you're like, all right, well, there just was a missile that went through an apartment block. What are you going to do?
我觉得不好。哦，我要评断一下。很不好。我说的是，就像你在哭一样。就像，你知道的，就像想象一下乌克兰就像说，“哦，我们不认为我们应该参与战争。我们不认为我们应该为任何防御事项做出贡献。” 然后你就说：“好吧，刚才有一个导弹飞进了一个公寓楼。你打算怎么办呢？”
It's like, oh, yeah, but we just don't think we should have anything to do with that. And it's just like, what a goddamn luxury to be able to sit in a place where you're like, I don't want to have anything to do with defensive freedom, right? Defensive the values, defensive human life against an oppressor coming in and coming across.
这个意思是，哦，是的，但我们觉得我们不应该与那个相关。就像，真是一个该死的奢侈，能够坐在一个地方，说：“我不想和防御自由有任何关系”，对吧？防御价值观，防御人类生命，抵制一个压迫者的入侵和侵犯。
So first of all, like that is such an IE view is that you don't need defense or you don't need to have technology as part of defense. So that's number one.
首先，就像IE观点一样，您不需要防御或不需要将技术作为防御的一部分。 这是第一点。
I think the value is guilty of being naive about what it means to be a superpower in the world. And whether you like it or not, you are going to need a defense structure, all right?
我认为价值观对于成为世界超级大国的意义有些天真。无论你喜不喜欢，你都需要防御机制，对吗？
The second bit is you don't always choose when you fight, but you choose with the technology that you build how you will respond or your capability of responding. So I think those are there.
第二部分是你并不总是选择什么时候打斗，但是你可以选择用你建设的科技来如何应对或者你应对的能力。所以我认为这些是存在的。
I think it hasn't proved a lot since the Maven space. That was 2019. I think it was 2019 was when that came. Yeah, and one of the things that changed was the administration. And I think people have become a lot more comfortable with the Biden administration and the Valley, despite the fact that Pentagon and intelligence agencies are not the same as the White House administration. But they've certainly become more comfortable.
我认为自从 Maven空间以来，并没有证明太多。那是在2019年。我想那是在2019年发生的。是的，其中一个变化就是政府治理。尽管五角大楼和情报机构并非白宫管理团队，但我认为人们已经对拜登政府和硅谷变得更加舒适。但他们肯定变得更加舒适了。
So I think part of that is being knocked down and I think people have realized and Landmore in Europe did a lot of that. Is it actually the Stuz Matter, right? And so that's come through. But if you're rewind to 2019, it was like, you had to have really clear thoughts and discussions with your employee base. At least they walk out and say, like, you know, this is not there. So part of that, I think, is understanding what it means to be in a world where war exists.
我认为其中一部分原因是我们被打击了，人们已经意识到欧洲的Landmore承担了很多责任。对吧，这实际上就是Stuz问题，这是显而易见的。但如果回到2019年，你必须与员工基础进行非常清晰的思想和讨论。至少他们说走就走，认为这不成立。我认为，其中一部分原因是要理解存在战争的世界意味着什么。
I think the second bit now, though, is not so much can Silicon Valley build those things. It's can they actually get into the hands of the soldiers and the people fighting on the front lines. And this is where we go back to procurement dynamics, right?
我觉得现在第二个问题不是硅谷能否建造那些东西，而是他们是否能够真正地送到士兵和前线作战人员手中。这就是我们回到采购动态的地方，是吧？
So we've gotten past the kind of, what I'd call social license to operate. It is acceptable to be a defense focused company. It's not ever going to be the mainstream kind of light building consumer apps. Is it going to be easier for you to recruit than it was three years ago? Yeah, 100%. Right? People come in and understand now that there's a mission that is actually important, right?
所以我们已经度过了我所谓的社会运营许可的阶段。成为一家防御性公司是可以接受的。它永远不会成为主流的轻型消费类应用程序。现在你们招聘人员会更容易吗？是的，百分之百。对吧？人们现在进来明白这是一项真正重要的任务。
And you know, I think one of the things here with defense is always to say, look, the work that you're doing is going to have an impact in a way that is literally life and death, right? And that, I think, is a mission that people want to get in and get behind. And it's acceptable now to do that, I think, three or four years ago, it was kind of like, you had to be very careful as you boast their subject.
你知道，我觉得防御方面的一件事情就是要说，你所做的工作，实际上会影响到生死存亡的程度，对吧？而且，我认为这是一项人们愿意参与的使命。现在这是可以接受的了，三、四年前还需要非常小心地宣传他们的项目。
I also think people have focused on problems that you have a luxury in the Valley of focusing on. Yeah, right? And it's just like war isn't something you think about in the Valley because it's just not present. Well, because we need a better app to walk our dogs. That's right. You better have to walk our dogs or whatever it is. And so, but it's also a long way away.
我也觉得人们关注的问题，在硅谷却是一种奢侈品。对吧？就好像战争在硅谷是不存在的，所以我们不需要考虑它。我们需要一个更好的应用程序来遛狗。你说的对。我们需要一个好的应用程序来遛狗或者处理其他的事情。但是这个问题也离我们很遥远。
Like the cultures of Washington, D.C. and Silicon Valley are a long way away. And you know, there's half a dozen flights a day that go back and forwards, but that only means that there's, you know, a few hundred people that make a commute between the two cities every day, which is tiny given the sort of the importance of those two places in our sort of trajectory as a country.
像华盛顿特区和硅谷的文化很遥远一样。你知道，每天有数十个航班来往，但这意味着每天只有几百人在两个城市之间通勤，这在我们国家的轨迹上来说微不足道，考虑到这两个地方的重要性。
But now I think that's improved. I think the bit that has to improve now is the ability to, I go through the procurement cycle, but I'd say B as well, the Pentagon has to also embrace the fact that they're not selling to large defense primes that can, you know, do cost plus and that can take forever and all the rest and they have to be able to move at the speed that the technology is moving at and move at the speed the companies are moving at behind it as well.
但是现在我认为情况有所改善。我认为需要改进的是能力，包括我能够顺利完成采购流程，还有像B公司一样，五角大楼也必须接受一个事实，那就是他们不是在向能够进行成本加成或需要很长时间才能完成的大国防公司销售，他们必须能够以技术发展的速度和背后的公司同步前进。
Not just because it's a good idea, it's because that's what your opponent's doing. You know, they're moving incredibly fast. In a way, the Pentagon has to become more like the App Store, right? Like if software and AI are eating defense, you have to be like, all right, well, we need an update next week, not three years from now. Well, you're, let's say, like you're in a war and you've got a computer vision system that's identifying your opponent.
不仅仅是因为这是个好主意，而是因为你的对手也在这样做。你知道，他们的行动非常快。某种程度上，五角大楼必须更像应用商店，对吧？如果软件和人工智能正在侵蚀国防，你必须像说，好的，我们下周需要更新，而不是三年后。假设你在战争中，你有一个计算机视觉系统，能够识别你的对手。
And they figure out a way to disguise themselves so that now your computer vision system doesn't work. Well, you don't wait three years to upgrade that. You want that in three minutes. So what's your system for procuring that next generation of capabilities as quickly as possible?
他们找到了一种方法来伪装自己，这样你的电脑视觉系统就无法工作。好的，你不会等上三年去升级它。你想要三分钟内做到这一点。那么你采购下一代能力的系统是什么，能够尽快地实现呢？
Is that happening? That overhaul that you're talking about that feels necessary? No, and America's got a problem, right? Because you can have all the technology, you can have technological advantage, you can even be winning the AI arms race. But if you can't get that in the hands of people that need to use it, then it's for naught, it's for nothing.
那真的会发生吗？你谈到的那个必要改革？不，美国面临一个问题，对吧？因为你可以拥有所有技术，你可以拥有技术优势，你甚至可以赢得人工智能军备竞赛。但如果你不能把它交到需要使用它的人手中，那么一切都白费了。
And I think we are a long way away from getting that done. And I think it's recognized as a problem, but I haven't seen anything over the last three or four years that have indicated that we've made substantive steps towards fixing it. There was what I would call a lot of innovation theatre that happened where people who like, here's $100,000 to try out this thing.
我认为实现这一目标还有很长的路要走。虽然人们已经认识到这是一个问题，但我在过去的三四年中没有看到任何迹象表明我们已经采取了实质性的步骤来解决它。有很多所谓的创新剧院发生了，在这些剧场里，人们会得到10万美元的资金以尝试某件事情。
And you see all of these well-meaning, but ultimately groups having very little impact of like, we're engaging tech. But none of that made into the hands of anyone who was actually fighting a war.. Have you talked about the last three or four years not seeing that change? What about the last three or four months?
你看到所有这些好心的组织最终都没有产生太大的影响，就好像我们正在利用技术一样。但是这一切都没有进入实际战争中的任何人手中。你有谈论过过去三四年没有出现这种变化吗？过去三四个月呢？
I either chat GBT effect. Yeah, so what's interesting now is you've got, I think people like opening their eyes and saying, wow, this came at it faster than we thought. So I think that's one. I think there's a realization that they need to do something now that these things are moving faster. But I don't think people know quite what that is.
我要么聊GBT效应，现在很有趣的是，我认为人们会睁开眼睛说：“哇，它比我们想象的更快地到来了。”所以我认为这是一个方面。我认为人们意识到，现在这些事情发展得更快了，他们需要立即采取行动。但我不认为人们知道该采取什么行动。
Right, what to do. Right, and for me, I think that the bit here, if I'm putting on my hat and saying, well, how do you solve for this? You don't, if you're saying you're in the hardware space because they understand hardware.
好的，我们该怎么办呢？我觉得对于我来说，如果我戴上帽子想解决这个问题，这里的关键在于硬件，如果你说你在硬件领域，他们会理解硬件，但是这个问题无法解决。
You're in the hardware space and you produce steel for ships. When a war starts, you don't say, hey, can you make us some steel? Yeah. You're like, well, we can, but we need to build a factory and et cetera. You actually go through and you say, we need you to be able to produce this much steel in case we should ever use it. And you have this kind of baseline of capacity for defense that allows you to manufacture. That needs to transfer across to software.
你们是硬件领域的，生产船用钢材。当战争爆发时，你们不会说：“嘿，你们能造些钢给我们吗？”是的，你们会说：“我们可以，但是我们需要建造工厂等。你们实际上要通过这个，你们要说，如果我们要使用的话，我们需要你们能够生产这么多的钢材。你们需要有这种防御储备能力的基准来进行生产制造。这需要在软件领域得到传递。
And what it says here is that, look, by the time you need it, you can't just spin up an entire software division, organization, whatever to produce the stuff. But until a war happens, which we hope never does, there's no need for it. So the way to get around this and the way we've done it in the past has said, we're going to build a baseline capacity that we can tap into.
这里说的是，当你需要它的时候，你不能只是在短时间内搭建整个软件部门或组织来生产所需的东西。但直到战争爆发，我们希望这不会发生，这些东西就没必要了。所以，我们过去的做法是建立一个基准能力，以便在需要时利用。
And so what we need is the software equivalent of that, which says, look, we're going to fund the production of X, Y, and Z. We hope never to have to use them. But we need to have that capability and people with the ability to do that stuff on tap should we ever need to. And that's, I think, the only way you can do this, right?
因此，我们需要一个软件版本的这个东西，它可以说，看，我们要资助 X、Y 和 Z 的生产。我们希望永远不需要使用它们。但是，我们必须拥有那种能力，并且有能力随时做出这样的东西，以备不时之需。我认为这是唯一的方法对吧？
Precuring things when a war breaks out is exactly the wrong way to do this. And so we sat down, for us, we had a whole bunch of stuff that could be deployed and having a big impact inside of Ukraine. This is like the kind of real-time intelligence. Real-time intelligence.
当战争爆发时预先处理事情，这样做实际上是错误的方法。因此我们坐下来，对我们来说，我们有一堆东西可以在乌克兰内部产生重大影响。这就像实时情报一样。实时情报。
For somebody like, I'm wherever, picks random spot in Ukraine. Yep. Being able to have common operating and architecture, situational awareness of what's unfolding from the multiplicity of data sources mediated by artificial intelligence, so you can make better common control decisions. Right. So we have that. We have the money allocated. The money's available, but there's no contract to put that on.
对于某人来说，我就像随机选定乌克兰的任意地点一样存在。嗯。通过人工智能中介多重数据源的情境意识，能够具有普遍的操作和架构，以便您能够做出更好的普遍控制决策。对的。所以我们有这个。我们拨款了资金。这些资金可用，但没有合同去实施。
And so you go through this kind of like thing, which is like, there's money, this technology, and there's need. And there's need. And then you go through, it's like, where's the contract? And so now, well, now to get a contract, here's the process, and here's how you unfold through it. Then you're in the line at the post office, basically. And so like 12 months later, you're there, but it's like, well, 12 months, like, that's not how these things work.
所以你要经历这种事情，就像有钱、有技术、有需求。还有需求。然后你开始了，就像在寻找合同。现在要签合同了，这是一个过程，你需要按照这个过程去进行。 最后你会排队在邮局里，但是这需要12个月后。但事实上，这些事情怎么可能只需要12个月呢？
So that's kind of the dynamic is this stuff on folds. I think the other bit here is like, if you make that simpler, you're going to attract more people to this problem, because people, as much as they're building dog walking apps, they don't really want to. No one gets up and says, you know what I want to do with my life is build dog walking app. I'm sorry for the dog walking app engineers out there, but really, like, you can fight for the ideological future of the planet.
所以这种动态就在于对褶皱的研究。我认为还有一个问题，就是如果你把它简化了，你会吸引更多的人来解决这个问题，因为人们不管是做狗遛app还是其他事情，都不是真正想要这么做的。我很抱歉针对那些狗遛app工程师，但是，你们真的可以为这个星球的意识形态未来而战。
Like, you can build dog. And then back to that, cool. He can build a dog walking app when you try. That's right. Once we've fought and won the ideological battle of the next century between Western liberal democracy and authoritarianism, then we can walk dogs. But it's like, everyone's going to choose that time and again, right?
就像你可以建造狗。嗯，很酷。他可以在你尝试时建造一个遛狗应用程序。没错。一旦我们打赢了下个世纪西方自由民主与威权主义的意识形态斗争，那么我们就可以遛狗了。但是，似乎大家总是会选择那个时机，对吧？
This is the same reason that people, when there was an invasion in Ukraine, they didn't line up to get out of the country. The young men lined up to get back into the country, because this is something that matters. And that story is as true in America as it is anywhere else in the Western world. But you've got to give that technology and the people building it space to go ahead and do it.
这就是为什么当乌克兰发生入侵时，人们没有排队逃离这个国家的原因，年轻的男子们排队回到了国家，因为这是重要的事情。这个故事在美国和西方世界其他地方同样真实。但你必须给予这项技术和建造它的人足够的空间可以继续前进。
And the reality is it's a tiny fraction of the price of any of the metal that you're building. And it's the future of where, like, conflict is going. And we honestly, we don't have a choice, but to engage in this and run incredibly fast and we have to win it. Because if we lose AI dominance to China, the offset kicks in and everything that we've built is destroyed.
现实是，它的价格只是你正在建造的任何金属的一小部分。它是冲突趋势的未来。我们必须加入到这项工作中并快速努力，必须赢得它。因为如果我们失去对中国的AI主导地位，偏移将发挥作用，我们建造的一切都将被摧毁。
So is the next weapon of mass destruction, the next kind of A-bomb equivalent, thinking about what this kind of embodiment of what that offset you're talking about? Is it like, I don't know, a thousand or a million, like, you know, $100 drones all loaded with just a little payload enough to kill one person and they just send them all, you know what I mean?
那么下一个大规模毁灭性武器，就相当于下一个原子弹，你考虑的是这种化身会给你谈到的那种平衡造成什么影响？它是像一千个或一百万只只装有足以杀死一个人的负载的小型$100无人机，然后它们就全部被发射，你知道我是什么意思吗？
Yeah, and this is sort of like this murder bots kind of dynamic, which Stuart Russell kind of has been banging on the drums.. I mean, but I'm chatting with him and, you know, and other folks that kind of share similar beliefs. And you know, look, I get it, right?
是的，这有点像是一种杀人机器人的动态，Stuart Russell一直在强调。但我正在和他以及其他分享相似信念的人聊天。你知道的，我明白这一点。
Like, you don't want to create an AI system that goes and destroys your population. But you certainly want an AI system that defends your population. And that will likely be a swarm of killer robots. They just will be aimed to kill the people that are attacking you. And so we're not going to escape autonomy in the weapon systems that we've got.
就像你不希望创造一个AI系统去摧毁你的人民一样，但你肯定希望有一个可以保护你的人民的AI系统。可能会是一群杀人机器人。它们只会瞄准那些攻击你的人。所以，我们无法避免我们拥有的武器系统中的自主性。
And indeed, it's even a false kind of thing should we bring autonomy? And we built in the Second World War, automatic anti-aircraft guns that tied to radar systems that improved the accuracy 10-fold by automating the targeting anti-aircraft guns, right? Like, we've had autonomy. We've had autonomy and precision weapons. What we're talking about is better autonomy.
我们如果引入自主性，这恰恰是一种虚假的选择吗？在第二次世界大战期间，我们建造了自动防空炮，与雷达系统联合起来，通过自动瞄准防空炮，使准确度提高了十倍。就像我们已经拥有了自主性和精确武器一样，我们现在讨论的是更好的自主性。
You know, I think there's no escaping autonomy. The question is, how is the opponent going to use it and how are we going to defend against that and how are we going to maintain dominance such that those conflicts never happen? But I think what does this mean if you get this, you know, orientated, you move from a space of building very complex single machines driven by humans into cheap, distributed swarm dynamics swarm intelligence controlled by AI systems.
你知道吗，我想自主性是无法逃避的。问题是，对手会如何利用它，我们将如何防御，并如何保持支配地位，以使这些冲突永远不会发生？但我想，如果你理解了这一点，你就会从由人驱动的非常复杂的单一机器构建的空间转移，进入由AI系统控制的便宜、分布式的群体动态群体智能的空间。
And it then says, how could you AI and how good is your low cost manufacturing to spend out tens of thousands of these things? So think of the F-35 versus 35,000 drones that emerge out and just kind of like go into its engine systems that get sucked through and destroy it. Or think of 10,000 C or underwater drones that come in at a battle group for an aircraft carrier, each one containing enough payload to destroy or cripple. And even if you knocked it one of them out, you're knocking each one of them out with a $400,000 missile or a $4 million missile. Right. So like this is where war is going to go. So like it's going to change that dynamic.
然后它说，你们的人工智能和低成本制造有多好，才能支出成千上万个这些东西？所以想象一下F-35和3.5万架无人机的区别，这些无人机会涌入发动机系统，并被吸入摧毁它。或者想象一下1万艘C型或水下无人机，在一支航母战斗群中进攻，每一个都包含能毁灭或瘫痪的足够有效载荷。即使你用一枚价值$400,000或$4,000,000的导弹击落了其中的一个，你也会破坏其中的每一个。所以这就是战争的未来，这将改变战争的动态。
But of course, the one there which we think a lot about, but we forget is there's no war unless there's a desire for war. And one of the bits that I think Ukraine has taught us is that I think Putin would have been right with his actions in his strategy, but for the fact that people, instead of turning around to walk out of the country, turned around and walked back in. If people didn't pick up, clash, and a carves to fight, if people around the world didn't pressure their governments to send javelins and tanks, there was no war to be had.
当然了，我们经常思考的一个问题是，如果没有战争之欲，就没有战争。我认为乌克兰给我们带来的重要启示是，普京的策略可能本来是正确的，但是人们不是选择离开这个国家，而是选择回到这个国家。如果人们不会挑起冲突，不会热衷于战斗，如果世界各地的人们不去向政府施压，要求派出飞弹和坦克，那么就不会有战争。
And so they won in Ukraine, they won the information war, they won the narrative, and they said, snake island and the woman with sunflowers in your pocket. And the ghost of Kiev, they won that, and there was a war to be fought. But without that spirit, there'd be no conflict.
所以他们在乌克兰获胜了，他们赢得了信息战，赢得了叙述权，他们说到了蛇岛和口袋里有向日葵的女人。他们赢得了基辅的幽灵，还有一场未打的战争。但是如果没有那种精神，就不会有冲突。
And if we look towards Taiwan, the single biggest thing that China wants is a conflict that looks like Hong Kong, which is like, there's nothing to see here. It's an internal Chinese matter. Don't engage versus something that looks like Kiev, which is like everyone decides that this is something they need to get in and support.
如果我们看向台湾，中国最想要的就是一场看起来像香港的冲突，那里似乎没什么值得关注的，它是中国内部事务，不要插手。相反，中国不想看到呈现出基辅的形象，即所有人都认为这是他们需要参与和支持的事情。
And so what you're going to see, and what you are seeing, is I think a concerted effort by China to win an information war. And the reason for that is if China can take Taiwan without kinetic destruction, it gets TSMC, or the semiconductor manufacturing, which controls 90 plus percent of all the advanced AI chips in the world. And it controls that piece of it.
所以你现在看到的，我认为是中国开展信息战的协调努力。原因是，如果中国能够在不使用动力破坏的情况下占领台湾，那么它就可以获得台积电，或半导体制造业，其控制了全球90多个百分点的所有先进AI芯片。它控制了这一部分。
And so if I'm China sitting here, I'm going to run a very, very concerted information operation. And I'd probably love to have a very popular application that every soldier and person had in their pocket, where I could control the information they can consume, and maybe distracted them or divided them or created information that said, hey, don't worry about Taiwan. It's an internal Chinese matter. And so they have no idea which is that.
如果我是中国，坐在这里，我会开展一场非常有力的宣传。我可能会喜欢拥有一款很受欢迎的应用程序，每个士兵和民众都会装在口袋里，可以控制他们可以消费哪些信息，或者分散他们的注意力，分裂他们的思想，或者制造一些信息说：“不用担心台湾，这是中国内部的事情。”所以他们不知道到底是哪种情况。
I'm just trying to think if there was an operation, where you could design that and put it in. I mean, wouldn't that be wonderful? And we're still dividing about whether or not TikToks are a good thing. Yeah, yeah. We live in bizarre times.
我只是在思考是否有一种操作方式，可以设计并放入那个东西。我的意思是，那不是很棒吗？而我们仍然分歧于TikToks是否是好事。是啊，是啊。我们生活在奇怪的时代。
Well, look, if I was a CIA, and I had the ability to implant an application that controlled the information stream of every single person in China to my liking, that would be considered the greatest intelligence when ever. And that is exactly what we're sitting in here today. And we're still having this debate of whether or not we should have TikTok in the pockets of soldiers and the pockets of politicians, of the pockets of literally every person on the age of 40.
听我说，如果我是中央情报局的人，并且有能力植入一个应用程序，控制中国每个人的信息流，让其按照我的意愿运转，那将是有史以来最伟大的情报成果。而今天我们正坐在这里，就是这个成果。我们仍然在争论是否应该让士兵、政治家甚至是四十岁以上的普通人将 TikTok 放在口袋里。
The truth is, with that in place, you can't fight a war in China or in Taiwan because no one's going to care.
事实上，如果出现这种情况，你在中国或台湾打仗都无济于事，因为没有人会在乎。
On that happy note.
关于这个开心的话题。
I care. I care.
我很关心。
But people are waking up.
但是人们正在觉醒。
But look, this is real.
但是你看，这是真实的。
We're going to be walking.
我们要去步行。
If we imagine we're walking into a world with a hot conflict versus China, how do we best prepare because the clock is ticking?
如果我们想象一下走进与中国的热点冲突的世界，我们该如何做最好的准备呢？因为时间正在流逝，我们必须要有所应对。
And everything from information superiority through to artificial intelligence, superiority through to the speed and engagement of the technology sector into that.
把从信息优势到人工智能，再到技术行业的速度和互动等方方面面都考虑进去。
All of this matters.
所有这些都很重要。
But this starts now.
但是现在就开始了。
It's not something we can think about in the future.
这不是我们未来能考虑的事情。
We are going to be in a fight for the dominant global military power of the planet.
我们将在争夺地球主导全球军事力量的战斗中参与。
And it's going to be America and the Western liberal democracies, or it's going to be a authoritarian China.
要么是美国和西方自由民主国家，要么就是威权主义的中国。
And I don't think that's a very hard decision from where I'm sitting.
我觉得，从我的角度来看，这不是一个很难的决定。
And if that's the thing that's in front of us, why is this not front and center for America and everyone who can make an impact and a difference?
如果这是摆在我们面前的事情，那为什么美国和所有可以产生影响和改变的人们不把这个问题放在首位呢？
Because at a time, bomb start dropping is too late.
因为在炸弹开始降落的时候，为时已晚。
Absolutely, especially when they're dropped by swarms of drones.
当它们被一大群无人机投下时，特别是那种情况，绝对是这样。
With an I.I. system that we can't counteract and dictate.
我们无法对抗和控制的I.I.系统。
That they have the opposite.
他们有相反的东西。
Right.
好的。
Well, that's why I wanted to have you on.
嗯，这就是为什么我想要邀请你参加的原因。
Appreciate it.
感谢你。
It's been fun.
这很有趣。
It's all good.
一切都好。
Thank you.
谢谢你。
And that was all the time we have.
我们的时间已经用完了，就这样。
I want to thank Sean for coming on.
我想感谢Sean的加入。
I actually have a meet to his office, which is really funny because there was literally nobody there.
实际上我去了他的办公室开会，这真的很搞笑，因为那里真的一个人也没有。
It was a Friday.
今天是星期五。
So we have the whole run of the place.
所以我们可以整个地方任意自由活动。
I want to thank you all for listening for the ratings, for the reviews, for telling your friends and neighbors about the pod.
我想感谢大家的聆听、评分、评论，以及向朋友和邻居介绍这个播客的支持。
Thank you, thank you, thank you.
谢谢你，谢谢你，谢谢你。
And that's it for me this week.
这就是本周我要说的全部了。
I was actually out in the field, so to speak, on a couple of different stories.
我实际上可以说是在现场，做了几个不同的故事。
So I'm kind of not sure what I'll be writing about this week.
所以我不是很确定这个星期我会写些什么。
It's been kind of a crazy few days for me.
最近几天有些疯狂，我有点不知所措。
But anyhow, please have a gander at thetimes.co.uk or you find me on Twitter at Danny Fortson where you can email me Danny.Fortson at SundayHive and Times.co.uk.
总之，请看看thetimes.co.uk，或者你可以在Twitter上找到我@Danny Fortson，你可以通过Danny.Fortson @SundayHive和Times.co.uk与我邮件联系。
That is it for me this week.
这就是这周的工作结束了。
Thank you as ever for listening.
谢谢你一如既往地倾听。
And we'll talk to you very, very soon.
我们很快就会和你交谈。
Bye bye.
再见。
Translation by DanNip.
None